James Wayman, San Jose
State University
Title:
Biometric Testing
Abstract:
Large-scale
biometric testing has a history of at least 25 years. However, each test
has used different testing and reporting protocols, making results very hard
to understand and causing longitudinal comparison of biometric device and
system performance to be impossible. Working Group 5 of the ISO/IEC JTC1
Standing Committee 37 on biometrics has been established to address these
variances in the hope that a single test and reporting standard can be
developed. But many within the field feel that a single "standard" might not
be possible. In this talk, we will review historical testing and reporting
protocols, point out the areas of controversy, and analyze in detail the
recent UK contribution to the SC37 WG5 process.
Josef Bigun,
Halmstad
University
Title: Multimodal Biometric
Authentication for Mobile Communication
Abstract: The
elements of multi-modal authentication along with system models are presented.
These include the machine experts as well as machine supervisors. These will
be contrasted to human performance. In particular fingerprint and speech based
systems will serve as illustrations of a mobile authentication application. A
signal adaptive supervisor, based on the input biometric signal quality, will
be discussed. Experimental results on data collected from mobile telephones
are reported demonstrating the benefits of the proposed scheme in mobile
communication systems. The presentation is based on these studies, for which
the research documented in has been instrumental.
Gary Strong, U.S.
Department of Homeland Security
Title: Current
Issues in Biometrics at the Department of Homeland Security
Organizer: Jonathon Phillips, NIST/DARPA
Panelists: Paul Griffin, Kevin Bowyer,
Doug Reynolds
Title:
Multi-Biometrics, Déjà vu?
Abstract:
Multi-biometrics is currently a hot
area of research in biometrics, pattern recognition, and computer vision.
Multi-biometrics is the study of combining results from multiple biometric
samples and multiple biometric algorithms. This includes (1) combining
multiple modalities; e.g., face and fingerprints; (2) multiple samples of the
same modality; e.g., five face images of the same person; and (3) multiple
algorithms on the same sample; e.g., fusing the results of three speaker
recognition algorithms on the same speech sample. All of these different
approaches show improvement in performance. Looking at all these results gives
the author a deja-vu feeling from face recognition in the 1990s. In the 1990s
most pattern recognition techniques demonstrated a reasonable ability to
recognition faces. It took a number of years to identify the most promising
approaches.
The key question
for multi-biometrics is how to determine the superior multi-biometrics
approaches. Sorting out the merits of different multi-biometrics is more
complex than with a single biometric modality. The first step in identifying a
superior approach is to separate out the effects of the modality, algorithms
for the individual modalities, and the fusion algorithm. Separating out these
effects is necessary so that one does not condemn a modality or combination of
modalities because of poor algorithms. This can be either a fusion algorithm
or individual modality algorithm. A second level of analysis considers the
scenario for using a multi-biometric system. Under different circumstances,
different multi-biometric configurations will be preferable. From a cost
perspective, a multiple sample approach will be most likely be cheaper than a
multiple algorithm approach which is likely to be cheaper than a multiple
modal approach. For robustness, a multiple modal approach is most likely to be
best. To move forward in multi-biometrics, these issues and others need to be
addressed. This panel will discuss these issues and raise others associated
with assessing the state of multi-biometrics, and paths for advancing
multi-biometrics.